Skip to content

[WIP] allow range to satisfy key distribution generally #23680

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
gene-bordegaray wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

[WIP] allow range to satisfy key distribution generally #23680

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
433 changes: 160 additions & 273 deletions datafusion/physical-expr/src/partitioning.rs
View file
Open in desktop

Large diffs are not rendered by default.

37 changes: 11 additions & 26 deletions datafusion/physical-optimizer/src/ensure_requirements/enforce_distribution.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ use std::sync::Arc;
use crate::output_requirements::OutputRequirementExec;
use crate::utils::{
add_sort_above_with_check, is_coalesce_partitions, is_repartition,
is_sort_preserving_merge, range_partitioning_satisfies_key_partitioning,
is_sort_preserving_merge,
};

use arrow::compute::SortOptions;
Expand Down Expand Up @@ -698,18 +698,13 @@ fn add_roundrobin_on_top(
}
}

// TODO: remove this temporary bridge once [`Partitioning::Range`]
// generally satisfies [`Distribution::KeyPartitioned`] through
// [`Partitioning::satisfaction`].
// <https://github.com/apache/datafusion/issues/23266>.
//
// Partial aggregates do not require key partitioning, but they preserve their
// input partitioning for the final aggregate. Until Range satisfies
// KeyPartitioned generally, this check keeps preserve_file_partitions from
// inserting RoundRobin between a reusable Range input and the partial aggregate.
fn partial_aggregate_preserves_reusable_partitioning(
// Partial aggregates require unspecified input distribution, but their output
// may already satisfy the final aggregate's key distribution because partial
// aggregation preserves/projects input partitioning. Keep that reusable output
// partitioning intact when preserve_file_partitions would otherwise insert
// RoundRobin below the partial aggregate.
fn partial_aggregate_output_satisfies_final_partitioning(
plan: &Arc<dyn ExecutionPlan>,
child: &Arc<dyn ExecutionPlan>,
allow_subset_satisfy_partitioning: bool,
) -> bool {
let Some(aggregate) = plan.downcast_ref::<AggregateExec>() else {
Expand All @@ -722,24 +717,15 @@ fn partial_aggregate_preserves_reusable_partitioning(
return false;
}

let group_exprs = aggregate.group_expr().input_exprs();
let output_partitioning = child.output_partitioning();
let eq_properties = child.equivalence_properties();
let key_distribution = Distribution::KeyPartitioned(group_exprs.clone());
let key_distribution = Distribution::KeyPartitioned(aggregate.output_group_expr());

output_partitioning
plan.output_partitioning()
.satisfaction(
&key_distribution,
eq_properties,
plan.equivalence_properties(),
allow_subset_satisfy_partitioning,
)
.is_satisfied()
|| range_partitioning_satisfies_key_partitioning(
output_partitioning,
&group_exprs,
eq_properties,
allow_subset_satisfy_partitioning,
)
}

/// Adds a [`SortPreservingMergeExec`] or a [`CoalescePartitionsExec`] operator
Expand Down Expand Up @@ -1308,9 +1294,8 @@ pub fn ensure_distribution(

let preserve_partial_aggregate_partitioning =
preserve_file_partition_threshold_met
&& partial_aggregate_preserves_reusable_partitioning(
&& partial_aggregate_output_satisfies_final_partitioning(
&plan,
&child.plan,
allow_subset_satisfy_partitioning,
);

Expand Down
59 changes: 1 addition & 58 deletions datafusion/physical-optimizer/src/utils.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,10 +18,7 @@
use std::sync::Arc;

use datafusion_common::Result;
use datafusion_physical_expr::{
Distribution, EquivalenceProperties, LexOrdering, LexRequirement, Partitioning,
PhysicalExpr, physical_exprs_equal,
};
use datafusion_physical_expr::{Distribution, LexOrdering, LexRequirement};
use datafusion_physical_plan::coalesce_partitions::CoalescePartitionsExec;
use datafusion_physical_plan::limit::{GlobalLimitExec, LocalLimitExec};
use datafusion_physical_plan::repartition::RepartitionExec;
Expand Down Expand Up @@ -161,60 +158,6 @@ pub fn is_repartition(plan: &Arc<dyn ExecutionPlan>) -> bool {
plan.is::<RepartitionExec>()
}

/// TODO: remove once Range generally satisfies KeyPartitioned requirements
/// through Partitioning::satisfaction.
/// See <https://github.com/apache/datafusion/issues/23266>.
///
/// Checks whether range partitioning satisfies a key partitioning requirement.
/// This is intentionally separate from general partitioning satisfaction while
/// range reuse is rolled out operator by operator.
pub(crate) fn range_partitioning_satisfies_key_partitioning(
partitioning: &Partitioning,
required_exprs: &[Arc<dyn PhysicalExpr>],
eq_properties: &EquivalenceProperties,
allow_subset: bool,
) -> bool {
match partitioning {
Partitioning::Range(range) => {
let partition_exprs = range
.ordering()
.iter()
.map(|sort_expr| Arc::clone(&sort_expr.expr))
.collect::<Vec<_>>();

if partition_exprs.is_empty() || required_exprs.is_empty() {
return false;
}

let eq_group = eq_properties.eq_group();
let normalized_partition_exprs = partition_exprs
.iter()
.map(|expr| eq_group.normalize_expr(Arc::clone(expr)))
.collect::<Vec<_>>();
let normalized_required_exprs = required_exprs
.iter()
.map(|expr| eq_group.normalize_expr(Arc::clone(expr)))
.collect::<Vec<_>>();

if physical_exprs_equal(
&normalized_required_exprs,
&normalized_partition_exprs,
) {
return true;
}

allow_subset
&& normalized_partition_exprs.len() < normalized_required_exprs.len()
&& normalized_partition_exprs.iter().all(|partition_expr| {
normalized_required_exprs
.iter()
.any(|required_expr| partition_expr.eq(required_expr))
})
}
_ => false,
}
}

/// Checks whether the given operator is a limit;
/// i.e. either a [`LocalLimitExec`] or a [`GlobalLimitExec`].
pub fn is_limit(plan: &Arc<dyn ExecutionPlan>) -> bool {
Expand Down
12 changes: 2 additions & 10 deletions datafusion/physical-plan/src/aggregates/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1934,7 +1934,7 @@ impl ExecutionPlan for AggregateExec {
}

fn input_distribution_requirements(&self) -> InputDistributionRequirements {
let requirements = InputDistributionRequirements::new(match &self.mode {
InputDistributionRequirements::new(match &self.mode {
AggregateMode::Partial | AggregateMode::PartialReduce => {
vec![Distribution::UnspecifiedDistribution]
}
Expand All @@ -1944,15 +1944,7 @@ impl ExecutionPlan for AggregateExec {
AggregateMode::Final | AggregateMode::Single => {
vec![Distribution::SinglePartition]
}
});
match &self.mode {
AggregateMode::FinalPartitioned | AggregateMode::SinglePartitioned
if !self.group_by.has_grouping_set() =>
{
requirements.allow_range_satisfaction_for_key_partitioning()
}
_ => requirements,
}
})
}

fn required_input_ordering(&self) -> Vec<Option<OrderingRequirements>> {
Expand Down
Loading
Loading